Prediction of pathologic stage in non-small cell lung cancer using machine learning algorithm based on CT image...

Area of Science:

• Radiomics and Machine Learning in Oncology
• Computational Pathology
• Medical Imaging Analysis

Background:

• Accurate pathologic staging is crucial for non-small cell lung cancer (NSCLC) treatment and prognosis.
• Current staging methods can be invasive and may not fully capture tumor heterogeneity.
• There is a need for non-invasive biomarkers to improve NSCLC diagnosis and staging.

Purpose of the Study:

• To identify imaging biomarkers from CT scans for NSCLC diagnosis and pathologic stage prediction.
• To evaluate the efficacy of multiple machine learning algorithms in analyzing CT image features.
• To assess the model's performance in predicting pathologic stage across different NSCLC subtypes.

Main Methods:

• Utilized CT image features from NSCLC patients (Stage IA-IV).
• Applied machine learning, including random forest for feature importance and SMOTE for dataset balancing.
• Validated the prediction model using training, testing, and external datasets (LUAD and LUSC from TCGA).

Main Results:

• A prediction model incorporating nine CT image features achieved high accuracy, precision, and recall.
• The model demonstrated strong performance on both training and testing datasets.
• External validation showed superior predictive accuracy for lung adenocarcinoma (LUAD) compared to lung squamous cell carcinoma (LUSC).

Conclusions:

• CT image features, analyzed by machine learning, can accurately predict NSCLC pathologic stage.
• The findings highlight the potential of radiomics as a non-invasive tool for NSCLC staging.
• Identified potential imaging biomarkers for NSCLC diagnosis and staging, with particular promise for LUAD.